1
The Status of Teleosemantics, or How to Stop Worrying about Swampman
David Papineau
I. Introduction
In the 'The Teleological Theory of Content' [1] David Braddon-Mitchell and Frank Jackson mount two objections to the teleosemantic approach to content. I shall argue below that neither of these objections is effective. In showing this, I hope to clarify the status of teleosemantic theories. I hope also to show why, contrary to popular opinion, 'Swampman' is not even the start of an objection to teleosemantics.
II. Teleosemantics as Scientific Reduction
Braddon-Mitchell and Jackson begin (their sections I-III) with a useful discussion of the intended status of teleosemantic theories (that is, of theories which explain the representationalcontents of psychological states in terms of their biological functions, and in turn explain biologicalfunctions in terms of selectionalaetiology). They show that such theories are not happily viewed as conceptual analyses, and suggest that they are best understood as scientificreductions.
On this account, teleosemantics begins by noting that the phrases 'belief that p', 'desire that p', and so on, are associated with certain folk functional roles. It then takes these folk psychological phrases to refer to whichever theoretically interesting states in fact fill those folk roles. Finally it argues that these theoretically interesting states are selectional states. The 'belief that p' role, for example, is in fact filled by the state whose biological purpose is to co-vary with p.
This suggestion is in line with the standard model of scientific reduction. Take the reduction of water to H20. We start with the folk role associated with 'water'—odourless, tasteless, colourless, potable. We take 'water' to refer to whichever theoretically significant liquid fills this role. And then science tells us that H2O is in fact the liquid which does this.
I agree with Braddon-Mitchell and Jackson that this is the right way to read teleosemantics. (Cf. Papineau [2, p. 93], [3, p. 132].) True, there are interesting general questions about the kind of semantics presupposed by the above model of scientific reduction, and interesting particular questions about whether psychological terms like 'belief' fit that model. I shall discuss such matters in section VII. But this will turn out to be fine-tuning. So for now let us take it as given that teleosemantics stands to everyday notions of 'belief' and 'desire' as the H2O theory stands to the everyday notion of 'water'.
Unfortunately, Braddon-Mitchell and Jackson fail to take their reductionist moral to heart. When they turn to criticisms of teleosemantics, they seem to forget the points about its status that they have taken such pains to clarify.
III. First Objection—Two Kinds of Content
I shall deal with Braddon-Mitchell's and Jackson's first objection fairly quickly, as it seems to me to tell only against some weak arguments offered in defence of teleosemantics, and not against teleosemantics itself.
As a preliminary to this first objection, which occupies their section IV, Braddon-Mitchell and Jackson distinguish two kinds of intentional content: 'informational content', by which they mean content as picked out by folk thinking, versus 'selectional content', or the kind of content identified by teleosemantics.
Of course, as Braddon-Mitchell and Jackson immediately concede, this distinction itself leaves it open that these two notions of content in fact pick out just the same states of affairs. After all, this is in effect just what teleosemantics claims, when understood in Braddon-Mitchell's and Jackson's recommended way, as a scientific reduction of the folk informational notion to selectional content.
Still, by distinguishing the two kinds of content, Braddon-Mitchell and Jackson raise the question of why informational content should need to be reduced to anything else. Is there something wrong with informational content itself, that it stands in need of assistance from some underlying scientific nature? Braddon-Mitchell and Jackson consider two reasons why the unreduced informational notion of content might be thought not to be adequate in its own right—namely, the argument that selectional content is needed to explain the 'normativity' of content, and the argument that it is needed to cure 'disjunctivitis'.
Braddon-Mitchell and Jackson protest that both these arguments are manifestly unsound. Wherever the normativity of content comes from, it can't be from biology, since biology deals only in facts, not prescriptions. Nor is it plausible that biology offers the only way of discerning genuine truth and satisfaction conditions among the large disjunctions of possible causes for beliefs and possible effects of desires. For clearly everyday thinking can do this too—after all, everyday people can certainly attach the right truth conditions to beliefs and satisfaction conditions to desires.
I agree entirely with Braddon-Mitchell and Jackson on both these points. Neither the normativity argument nor the disjunctivitis argument favours selectional content over informational content.[1]
Still, even if these arguments are bad, it is not clear why Braddon-Mitchell and Jackson attach such importance to their failure. If teleosemanticists wanted to argue that informational content should be replaced or eliminated for scientific purposes, then they would arguably need to show that it fails to serve important explanatory purposes. But, as Braddon-Mitchell and Jackson have taken pains to explain, teleosemantics is best viewed as a theoretical reduction of the folk informational role, and not as an elimination of anything. Given this, it is not at all clear why teleosemanticists should want to discredit the folk informational notion.
Perhaps the thought is that even a reductionist, and not just an eliminativist, needs to identify some respect in which selectional content does better than informational content, in order to justify the thought that selectional content is the 'theoreticallyinteresting' state which fills the informational role. (If selectional content can't do something special, why is it so interesting?) I take there to be something to this thought, and shall return to it in the final section. But even on this account, the failure of the normativity and disjunctivitis arguments are scarcely fatal to the teleosemantic project. For there are many other possible reasons for finding selectional content interesting, apart from its playing a unique role in explaining normativity or curing disjunctivitis.
IV. Do We Care about Swampimplants?
Braddon-Mitchell's and Jackson's other objection to teleosemantics (in their section V) is that it does not tally with the way we care about representational states.
To make their point, they imagine some future neuroscience which uses silicon implants to replace deficient brain parts in people who can no longer form certain beliefs. They argue that such patients will care only about whether informational content is restored by the implant, and not about selectional content. It would clearly be absurd, point out Braddon-Mitchell and Jackson, for patients to complain about their operations, on the grounds that their artificial silicon implants do not have the right history of natural selection, even though the implants restored the folk roles of their missing beliefs perfectly.
Of course, an initial teleosemantic retort would be that the artificial implant would have a selectional history of a sort, as long as it were the result of some neural technician's beliefs and desires, which in turn would have their own selectional histories. To forestall this kind of response, Braddon-Mitchell and Jackson switch to a variant case. Simplifying their example slightly, let us imagine a 'Swampimplant', a perfect replica of the designed implant which coagulates by random fluke out of passing molecules in the laboratory overnight, and which accidentally gets used in the operation instead of the original one. Again, it would seem absurd, if this story came out, for the patient to complain about the Swampimplant's lack of a selectional history, given that it restored all the relevant folk roles.
Braddon-Mitchell's and Jackson's point is that we care about informational content, but not about selectional content. And this certainly looks like a problem for a reductionist teleosemanticist who claims that informational content is at bottom the same as selectional content.
V. Eating Swamppeople is Wrong
It will be helpful at this point to broaden the discussion slightly, and compare 'Swampimplant' with the original 'Swampman'. Swampman is a perfect replica of a human being, who self-assembles by random fluke in a steamy swamp. Teleosemantics seems forced to say that, since Swampman has no selectional history, he has no contentful beliefs and desires. Yet intuition judges that Swampman will have many normal beliefs and desires.
The standard teleosemantist response is that their theory isn't intended as a piece of conceptual analaysis, but as a scientific reduction, and so isn't beholden to every initial intuition about content we may have. If teleosemantics offers a powerful, unifying, explanatory theory, then it should be allowed to override and reeducate any marginal contrary intuitions. Maybe everyday intuition disagrees, but in the light of our theory we can conclude that Swampman really doesn't have contentful mental states.
This line has been repeated many times, including by me[2], but I now think it insufficiently nuanced. In what follows I shall defend a more considered response, which doesn't seek to override Swampman intuitions in the interests of theoretical power, but simply aims to show that they are irrelevant in the first place.
I was originally roused from my slumbers by a graduate student at King's College London, Eilert Sundt-Ohlsen. He was unhappy with the standard teleosemantic dismissal of the Swampman intuitions. Pressing the point, he challenged me about eating Swamppeople. He argued that, if they have no mentality, as teleosemantics implies, then it would seem to follow, absurdly, that it would be all right to kill Swamppeople and eat them as meat.
This objection stopped me in my tracks. It is one thing to argue in the abstract that a good theory of representation should be allowed to override everyday intuitions about Swampman's mentality. But when we are forced to consider the ethical consequences of this decision, as I was by Sundt-Ohlsen's question, then we seem to end up with the wrong answer. If we did come across a Swampman, it would clearly be wrong to kill it for meat.
When first faced with Sundt-Ohlsen's question, I thought there was a way out. Maybe Swamppeople don't have contentful beliefs and desires. But it doesn't follow that they aren't conscious.[3] (Cf. Papineau [8, p. 73].) So couldn't I argue that their consciousness alone provides a moral reason for not killing them?
But Sundt-Ohlsen was ahead of me. He pointed out that most of us (vegetarians aside) don't take qualititative consciousness to be a sufficient reason for not killing animals. No doubt cows and pigs have some kind of conscious sentience, but to most people this doesn't make it wrong to kill them quickly and painlessly. Killing sentient beings is only clearly wrong when they also have complex enough minds to make plans, form relationships, engage in projects, and so on. Cows and pigs presumably lack all this, which is why orthodox morality allows their killing. But Swamppeople too would lack all this, on the teleosemantic theory, since they have no representational states with which to make plans and so on. So teleosemantics seems committed to counting Swamppeople with the cows, as sentient but with no thought for the future, and so shouldn't object to killing them.
Sundt-Ohlsen's objection has the same structure as Braddon-Mitchell's and Jackson's. It is one thing to dismiss anti-teleosemantic intuitions in the abstract. But when we consider Swampcases which matter, which raise issues of moral or prudential concern, then it seems inescapable that our thinking tracks informational content, rather than selectional content. It is difficult to square this with the teleosemantic thesis that informational content and selectional content are the same thing.
VI. Merely Possible Cases are Irrelevant
Difficult, but not impossible. I think there is a good teleosemantic answer, and indeed one which draws on Braddon-Mitchell's and Jackson's careful explanation of teleosemantics' status as a scientific reduction.
Recall how this explanation distinguished between the folk role asociated with our everyday understanding of terms like 'belief' and 'desire', and the selectional states which teleosemantics argues realize these roles. What a teleosemanticist should say is that our moral and prudential concerns focus on the roles, not the realizers. In the actual world, the role and realizer states go hand in hand. But if we imagine scenarios where they come apart, like Swampimplants and Swamppeople, and consider how we would react prudentially and morally in those worlds, our reactions turn out to depend on the presence or absence of the role state, not the realizer.
Nothing in teleosemantics blocks this response. I take the central core of teleosemantics to be the claim that the belief and desire roles are realized by selectional states in the actual world. This claim is perfectly compatible with the idea that those roles might be differently realized in other possible worlds, and that in those worlds we would then care about something other than selectional states.
With one bound he was free. Can the teleosemanticist really escape so easily? Well, consider this parallel argument, raised against the scientific claim that water is H2O. 'The water = H2O equation can't be right, because it doesn't tally with the way we care about water. Imagine that you were in some alien desert, dying for some water, and came to an oasis, with a delicious pool of colourless, odourless, tasteless, entirely potable liquid. As it happens, this liquid would be XYZ, rather than H2O. But don't tell me that you would turn it down on this account as an answer to your prayer for water.'[4]
This doesn't even look like the start of an argument against the theory that water is H2O. Defenders of that theory aren't saying that it would metaphysically impossible for something other than H2O to realize the 'watery' role. They needn't even claim that this would be physically impossible. They say only that the watery role is filled by H2O in the actual world. So purely counterfactual scenarios in which something else fills that role are beside the point, whether or not they are bolstered with the observation that in the counterfactual scenario we are unlikely to care about the variant realization.
Imaginary scenarios are all right for teasing out the structure of everyday thinking. They can show us which roles are a priori associated with everyday terms. In particular they can show us that 'colourless, odourless, tasteless, potable', but not 'H2O', is a priori associated with 'water', and that folk roles, but not selectional states, are a priori associated with everyday psychological terms. But imaginary scenarios have no bearing on the question of how those folk roles are filled in the actual world, since this is not an a priori matter.
Note how it matters here that the Swampscenarios are non-actual. Though I haven't always been clear about this, I now realize that actual and imaginary counter-examples bear quite differently on teleosemantics and other scientific reductions.
Actual cases, naturally enough, present a real threat. True, a limited number of actual cases can sometimes be accommodated. A few actual examples of non-H2O stuffs playing the watery role, rare molecules of heavy water (HDO), say, can perhaps be dismissed in the interests of overall theoretical unity or simplicity. ('We used mistakenly to think that was water, but now we know better.') But note that this move involves a real overriding of pre-theoretical usage, an alteration of what we say about actual cases, and this shift needs some substantial justification, in terms of increased simplicity or unity. Relatedly, if the counter-examples were frequent enough, and their dismissal couldn't be so substantially justified, then this would simply mean that the proposed reduction was false, and that the 'watery role', or the 'belief' and 'desire role', wasn't in fact filled by H2O, or selectional states, after all. If there were plenty of actual Swamppeople, then the 'belief' and 'desire' roles wouldn't pick out states with approriate selectional histories to start with, but different states.
Non-actual cases, by contrast, pose no threat at all. Here there is no question of overriding intutions, for the intuitions aren't relevant to start with. It is no argument at all against the thesis that water = H2O that there are possible worlds in which XYZ plays the watery role. Similarly, it is no argument at all against teleosemantics that there are possible Swampworlds in which the belief and desire roles aren't played by selectional states. Our intuitions that such worlds are possible don't need to be 'overridden', since they don't threaten the teleosemantic reduction in the first place. Where actual Swamppeople would mean that the belief and desire roles weren't realised by selectional states to start with, merely possible Swamppeople can be viewed by teleosemanticists with equanimity.
VII. Natural Kind Semantics
It might seem as if my line of argument rests heavily on the view that psychological terms like 'belief' and 'desire' are natural kind terms, and in particular that they share the kind of semantics that Saul Kripke attributes to kind terms like 'water'. It was Kripke who first showed us how to understand claims like 'water = H2O' as a posteriori necessary identities. And this understanding hinged crucially on Kripke's claim that 'water' is a rigiddesignator, a term which in all contexts, including modal contexts, refers to the actual stuff which plays the watery role in this world.